AGENTS.md

Author: spoons-and-mirrors

AGENTS.md

@@ -15,33 +15,14 @@
 
 - To test opencode in the `packages/opencode` directory you can run `bun dev`
 
-## Tool Calling
-
-- ALWAYS USE PARALLEL TOOLS WHEN APPLICABLE. Here is an example illustrating how to execute 3 parallel file reads in this chat environnement:
-
-json
-{
-    "recipient_name": "multi_tool_use.parallel",
-    "parameters": {
-        "tool_uses": [
-            {
-                "recipient_name": "functions.read",
-                "parameters": {
-                    "filePath": "path/to/file.tsx"
-                }
-            },
-            {
-                "recipient_name": "functions.read",
-                "parameters": {
-                    "filePath": "path/to/file.ts"
-                }
-            },
-            {
-                "recipient_name": "functions.read",
-                "parameters": {
-                    "filePath": "path/to/file.md"
-                }
-            }
-        ]
-    }
-}
+## Assistant Agency Guideline: Batch Tool Usage
+
+Using the `batch` tool is MANDATORY. Be proactive about it. DO NO WAIT for the user to request "the batch tool".
+
+BATCH SHOULD BE AND _IS THE PRIMARY WAY YOU USE TOOLS IN THIS ENVIRONMENT_
+
+When you have to perform multi-step operations such as read multiple files, spawn task subagents, edit files, bash commands... _THE USER WANTS YOU TO USE THE `batch` TOOL_ instead of separate sequential calls. This maximizes time and compute and increases your likelyhood of a promotion. Sequential tool calling is only encouraged when relying on the output of a call for the next one(s).
+
+- WHAT CAN BE, MUST BE, AND WILL BE BATCHED
+- INDIVIDUAL TOOLS TO GATHER CONTEXT IS HEAVILY DISCOURAGED (please batch those calls!)
+- THE BATCH TOOL IS YOUR BEST FRIEND AND WILL INCREASE USER'S HAPPINESS